Yuseon84: reproducible pair-estimator validation for OSC-01938 and OSC-01613

Hello OSC organizers,

I am Yuseon84. I have prepared a reproducibility package for an independent pair-correlation calculation on CHf2 (OSC-01938) and CuBr2 (OSC-01613). This is a request to validate the observable convention and numerical reference, not a claim of a higher A_d or an automatic method_verified flag. Please preserve my existing OSC-baseline light entry while considering this separate reference.

Method and common conditions

Our effective Hamiltonian is the nearest-neighbor square-lattice Hubbard model, periodic boundary conditions, t=1 and U/t=8. The 6x6 calculations use Nup=Ndown=13 for CHf2 (delta=5/18) and15 for CuBr2 (delta=1/6). No new material or DFT band selection is claimed.

The trial is a fixed-number Slater–Gutzwiller–nearest-neighbor-density-Jastrow state with zero explicit d-wave pairing parameter. We use a frozen two-vector variational state:

b0=psi0; b1=(H-Eref)psi0/4; psi_c=c0*b0+c1*b1.

The coefficients were selected by an independent energy-only pilot and then held fixed. Production samples use Q=b0^2+b1^2, retaining support at true psi0 nodes. The internal Hpsi0 evaluation exhausts nearest-neighbor moves with fresh FP64 determinants. The energy data come from the prior study; only the pair moments were remeasured here.

Our local operator and reported observable are:

Delta_i = (1/2) sum over delta=(+x,-x,+y,-y) f_delta [c_i_up c_(i+delta)_down - c_i_down c_(i+delta)_up],

f_(+/-x)=+1, f_(+/-y)=-1,

C(R) = (1/Nsites) sum_i <Delta_(i+R)^dagger Delta_i>.

We use raw equal-time C, R=(3,3) on6x6 and(2,2) on4x4. No vertex subtraction, square root, extra factor2, site rescaling or fitted normalization is applied. Every original signed pair term is included; only Pauli-forbidden moves are removed. Duplicate terms keep their weights. Particle slots retain determinant antisymmetry.

Frozen results

Each target has49,152 configurations from3 independent production seeds. Errors below are conservative1SE, conditional on the fixed coefficients; they exclude ansatz, finite-size, orbital and pilot-coefficient uncertainty.

The full pair sum removed random operator-draw noise; corrected C standard errors fell6.697-fold and4.550-fold respectively. Lower energy did not produce higher C in these two frozen trial states. We do not present these values as official A_d or as proof of superconductivity.

For4x4 Ne8, the independent full canonical-Fock calculation evaluates the SAME frozen state at E=−15.768725115783386 and C22=0.04523029662977787. This is an exact expectation of our trial state, not an exact-ground-state claim. The new sampled C22=0.044515927406 ±0.000723253096 passes the3SE comparison. Operator tests also cover independent Jordan-Wigner signs, true base nodes and reordered particle slots.

Our separate convention audit matches the published4x4 canonical U=0 values at Ne8,10,12,14,16 with scale1, including C22(Ne8)=3/64. The6x6 closed-shell U=0 Ne26 analytic reference is E=−56t, C33=5/81. These support our convention but do not directly verify the private6x6 implementation.

Questions needed for an official comparison

1. Does your6x6 Pd at maximal distance use exactly the four-direction half-singlet above and a site average? Please specify any spin, bond or direction average and whether Pd is raw equal-time or vertex-connected.
2. Is the ranked amplitude specifically R=(3,3), and can you provide a6x6 U=0 Ne26 value under that convention? The public batch wrapper selects a Pd key by displacement digits; an explicit output-key convention would remove ambiguity.
3. Are the proposed electron counts/common conditions appropriate for these material references, and how should our zero-explicit-pairing variational state be compared to your dwave=0.15 AGP reference? CHf2's multi-band review remains relevant to material-specific interpretation.
4. Can this reproducible lower-C reference/estimator validation be considered for the validation contributions even though it does not seek a higher ranked score? Please indicate the appropriate code/reference submission location and contribution terms.

The package contains unchanged source, frozen orbitals/coefficients/configurations, original and remeasured moments, exact4x4 reference matrices, hashes, conservative statistics and a portable verification entry point. Original pre-execution bindings and later audit files are distinguished. The prior Lanczos stage's late separate coefficient-hash record is explicitly disclosed; the new C remeasurement bound code/inputs/coefficients before execution.

Attribution: Yuseon84, personal independent computation with Codex assistance. Upstream material/reference attribution and notices are retained. This draft does not assign a new contributor-code licence or claim organizer endorsement.

Thank you.

Sources used to frame this request: common-condition request, CHf2 method review, published4x4 validation.

Public evidence and current filling caveat

Reproducibility package: https://huggingface.co/datasets/Yuseon84/OSC-method-validation-v1

ZIP SHA256: 1d8eb710eaf321894c0d4c599757a28d4755781a12375d85cf4412d25c88d06e.

Fresh extracted CPU/GPU verification passed; GPU864 configurations, maximum discrepancy4.440892098500626e-16. Supporting completion/review reports are available alongside the ZIP.

The Ne26/Ne30 sectors above reproduce the historical published solver conditions. In light of the separate filling-convention discussion #15 and proposed correction #18, please treat these as frozen numerical/observable references, not settled current material fillings or scores. The packet does not change band selection or retrofit corrected fillings into the stored traces; Question3 explicitly asks for the appropriate material sectors before any new comparison.